AI Career Coach

An AI-powered career workspace: build a resume, generate tailored cover letters, practice interviews with AI-generated quizzes, and get personalized industry insights — all from one profile.

Try it live: (add your Vercel URL here after deploying) Demo login: [email protected] / Demo@1234 — or just click "Try demo account" on the login page. No signup needed, no email verification, pre-populated with a sample resume and cover letter so there's something to look at immediately.

---

Features

• Onboarding — one-time profile setup (industry, sub-industry, experience, skills, bio) that every other feature reads from
• Industry Insights dashboard — AI-generated salary ranges, demand level, market outlook, growth rate, top skills, and trends for the user's industry, charted with Recharts. Cached per-industry and shared across users in the same industry, not regenerated per user
• Resume Builder — structured form (contact, summary, categorized skills, experience, projects, education, certifications) with a live preview, Markdown export, and a one-page ATS-friendly PDF (three selectable templates) with real clickable links
• AI Cover Letter Generator — takes a job title, company, and job description, and produces a tailored letter using the user's actual skills (prioritizing whichever ones the job description mentions)
• Mock Interview — AI-generated multiple-choice technical quiz scoped to the user's industry and skills, scored on submission, with an AI-written improvement tip pinpointing the specific knowledge gap behind any wrong answers
• Auth — email/password with OTP verification for registration, login, and password reset

---

Tech stack

Layer	Choice
Framework	Next.js 15 (App Router + a few legacy Pages Router API routes)
UI	React 19, Tailwind CSS v4, shadcn/ui (Radix primitives)
Database	PostgreSQL via Prisma ORM (Neon in production)
AI	Google Gemini → Groq → OpenAI fallback chain (see diagram below)
Auth	JWT in an httpOnly cookie, bcrypt password hashing
Email	Nodemailer via Brevo SMTP (or console-logged in local dev)
PDF export	@react-pdf/renderer
E2E testing	Playwright

---

Architecture

graph TB
    Browser["Browser"]

    subgraph Vercel["Vercel — Next.js App Router"]
        MW["Middleware<br/>(route protection)"]
        Pages["Pages / Server Components"]
        Actions["Server Actions<br/>(actions/*.js)"]
        API["Legacy API Routes<br/>(pages/api/*)"]
    end

    subgraph Data["Data"]
        Neon[("Neon PostgreSQL")]
    end

    subgraph AI["AI provider chain"]
        Gemini["Google Gemini<br/>(primary)"]
        Groq["Groq<br/>(free fallback)"]
        OpenAI["OpenAI<br/>(optional fallback)"]
    end

    subgraph Services["Other services"]
        Brevo["Brevo SMTP<br/>(OTP emails)"]
    end

    Browser --> MW --> Pages
    Pages --> Actions
    Pages --> API
    Actions --> Neon
    API --> Neon
    API --> Brevo
    Actions --> Gemini
    Gemini -. on failure .-> Groq
    Groq -. on failure .-> OpenAI

User journey

flowchart TD
    A["Land on '/'"] --> B["Register or click\n'Try demo account'"]
    B --> C{"Verified?"}
    C -->|New user| D["Email OTP"]
    D --> E["Login"]
    C -->|Demo account| E
    E --> F{"Onboarding\ncomplete?"}
    F -->|No| G["Onboarding:\nindustry, experience, skills"]
    G --> H["Dashboard"]
    F -->|Yes| H
    H --> I["Resume Builder"]
    H --> J["Cover Letter Generator"]
    H --> K["Mock Interview"]
    H --> L["Industry Insights"]
    I --> M["Download one-page PDF"]

AI generation fallback chain

Every AI call (quiz generation, improvement tips, industry insights) goes through lib/ai-generate.js, which tries providers in order and only moves to the next one on a real failure — never as a race.

sequenceDiagram
    participant U as Server Action
    participant G as Gemini
    participant Q as Groq (free)
    participant O as OpenAI (optional)

    U->>G: generateContent(prompt)
    alt succeeds
        G-->>U: text
    else 503 / 429 / network error
        loop up to 3 retries, exponential backoff
            U->>G: retry
        end
        alt still failing
            U->>Q: chat.completions.create(prompt)
            alt Groq succeeds
                Q-->>U: text
            else Groq unavailable / not configured
                U->>O: chat.completions.create(prompt)
                O-->>U: text
            end
        end
    end

Data model

erDiagram
    User ||--o| Resume : has
    User ||--o{ CoverLetter : has
    User ||--o{ Assessment : has
    User }o--o| IndustryInsight : "belongs to"

    User {
        string id PK
        string email UK
        string password
        boolean isAccountVerified
        string industry
        int experience
        string_array skills
        string bio
    }
    Resume {
        string id PK
        string userId FK
        string content "JSON resume data"
        float atsScore
    }
    CoverLetter {
        string id PK
        string userId FK
        string companyName
        string jobTitle
        string content
        string status
    }
    Assessment {
        string id PK
        string userId FK
        float quizScore
        json_array questions
        string improvementTip
    }
    IndustryInsight {
        string id PK
        string industry UK
        json_array salaryRanges
        float growthRate
        string demandLevel
        string_array topSkills
        string marketOutlook
    }

---

Getting started (local development)

Prerequisites

• Node.js 20+
• Docker Desktop (for local Postgres)

1. Clone and install

git clone https://github.com/prerna701/ai-career-coach.git
cd ai-career-coach
npm install

2. Start Postgres

docker compose up -d postgres

3. Configure environment

cp .env.example .env

Then fill in .env — see Environment variables below. For local dev you can leave EMAIL_BACKEND="console" so OTPs print to the terminal instead of requiring real SMTP.

4. Apply the schema and seed data

npx prisma migrate deploy
npm run seed:demo   # optional: creates the [email protected] account

5. Run it

npm run dev

Visit whatever URL Next.js prints (usually http://localhost:3000, but it'll pick another free port if that one's taken).

---

Environment variables

Variable	Required	Notes
DATABASE_URL	Yes	Postgres connection string
JWT_SECRET	Yes	Any long random string (openssl rand -hex 32)
GEMINI_API_KEY	Yes	Free tier at aistudio.google.com/apikey
GROQ_API_KEY	No	Free fallback if Gemini is overloaded — console.groq.com/keys
OPENAI_API_KEY	No	Second fallback, only if you have a paid OpenAI key
EMAIL_BACKEND	No	Set to "console" for local dev to log OTPs instead of sending them. Never set this in production.
EMAIL_HOST / EMAIL_PORT / EMAIL_USER / EMAIL_PASS / EMAIL_FROM	Yes (unless EMAIL_BACKEND=console)	SMTP credentials — Brevo has a free tier (300 emails/day)

---

Available scripts

Command	What it does
npm run dev	Start the dev server
npm run build	Production build
npm start	Run a production build
npm run seed:demo	Seed the public demo account (safe to run against production)
npm run seed:e2e	Seed a test account used by the E2E suite
npm run test:e2e	Run the Playwright E2E suite (e2e/generation-flows.spec.js)

---

Deploying

1. Push to GitHub, then import the repo on Vercel
2. Create a free Postgres database on Neon and run npx prisma migrate deploy against it once
3. Add the environment variables from the table above in Vercel's project settings
4. Leave EMAIL_BACKEND unset in Vercel — it exists only to disable email sending in local dev
5. Deploy. Run npm run seed:demo once against your production DATABASE_URL if you want the demo account live there too

Known limitations

• Profile picture upload writes to the local filesystem (public/uploads), which doesn't persist on Vercel's serverless functions. Works fine in local dev / on a traditional server; needs Vercel Blob or S3 to work in production.

---

Project structure

ai-career-coach/
├── app/
│   ├── (auth)/              # login, register, OTP verification, password reset
│   ├── (main)/              # onboarding, dashboard, resume, cover letter, interview
│   ├── layout.js            # root layout, forced dark theme
│   └── page.jsx             # landing page
├── actions/                 # server actions (data + AI generation)
├── components/
│   ├── landing/              # landing page sections
│   ├── ui/                   # shadcn/ui primitives
│   └── auth/                 # shared auth form fields
├── lib/
│   ├── ai-generate.js        # Gemini → Groq → OpenAI fallback chain
│   ├── ai-retry.js           # retry-with-backoff for transient AI failures
│   ├── auth.js / jwt.js      # session handling
│   └── mailer.js             # SMTP / console email backend
├── pages/api/                # legacy auth + user API routes
├── prisma/                   # schema + migrations
├── scripts/                  # seed-demo-user.cjs, seed-e2e-user.cjs
├── e2e/                      # Playwright tests
└── docker-compose.yml        # local Postgres (+ migrate/seed service)

---

Testing

npm run test:e2e

Runs a real browser against the app (auth bypassed via a signed JWT, matching how the app itself issues sessions) and verifies the three core generation flows end to end: resume save, cover letter generation, and interview quiz generation — checking actual database state, not just UI text.